#38 · ML / AI methods

Model-based inference of transport and tissue structure from optical imaging data

The items share a computational inference strategy in which estimation is built on explicit structural constraints or acquisition metadata. The serial OCT preprint estimates microscale fiber orientation distribution functions using multiscale Frangi filtering and anatomically constrained particle filtering, validated on FiberCup. The Multiview-SPIM-micro PIV pipeline recovers time-resolved, three-component, three-dimensional blood flow in the beating zebrafish heart by combining cardiac phase assignment with z-stack processing. The flow matching study applies a generative approach to segment 3D curvilinear vessels in medical images and reports gains over baselines on three public datasets. The set is heterogeneous in modality and scale, so the thread is dominant rather than universal. What is new is the explicit integration of anatomical or temporal structure into estimators for transport and tissue geometry.